COMMENTARY
Caste and Census: A Forward
Looking Strategy
Sonalde Desai
In modern India, vast quantities
of research have documented
caste-based inequalities in many
dimensions of well-being.
If these inequalities are not
simply imagined but reflect social
processes that deserve public
policy attention, incorporating
questions about caste in the
census is imperative. However,
there is a need to devise an
accounting framework that has
clarity of purpose since there are
many complexities involved in
collecting caste data.
Sonalde Desai (sonalde.desai@gmail.com)
is with the National Council of Applied
Economic Research, New Delhi, and the
Department of Sociology, University of
Maryland, College Park, MD, USA.
10
O
pponents of inclusion of caste in
the census argue that for a society
which seeks to abolish caste-
based inequalities, a census that inquires
about caste identities is a retrogressive step
since it is more likely to solidify caste-
based divisions than to obliterate it. Fol-
lowing Benedict Anderson’s argument that
censuses played an important role in crea-
tion of imagined communities that tran-
scend face-to-face associations (­Anderson
1983), many scholars have argued persua-
sively that colonial censuses created caste
as enumerated communities (Das 2003)
and solidified hitherto fluid identities
(Dirks 2001). Resistance to this reification
of social difference often emerges in the
form of reluctance to collect caste data. In
many ways, this reluctance is similar to
that observed in collection of racial statis-
tics in other cultures (Zuberi 2001).
While there is some justification to this
argument, we are now living with the
a­ftermath of these political processes. In
modern India, vast quantities of research
have documented caste-based inequalities
in many dimensions of well-being, includ-
ing income, education, health and access to
employment (Govinda 2002; Thorat and
Newman 2009; Desai et al 2010; Desh-
pande 2000). If these inequalities are not
simply imagined but reflect social processes
that deserve public policy attention, then
incorporating questions about caste in cen-
sus is imperative. However, it is easier to
suggest that caste be counted (e g, the EPW
editorial of 22 May 2010), than to devise an
accounting framework. Much of the diffi-
culty emerges from lack of clarification re-
garding the purpose of this accounting.
Why Collect Caste Data?
The most recent demand for a count of the
Other Backward Classes (OBCs) has come
from a powerful OBC lobby that hopes for
an increase in OBC reservations if the
july 17, 2010  vol xlv no 29  EPW   Economic & Political Weekly
count turns out to be higher than expect-
ed. The 27% reservation for OBCs is based
on the estimate by the Mandal Commis-
sion that OBCs form about 52% of the pop-
ulation and since all OBC families are not
poor or “backward”, a quota limit set at
about half the estimated population
makes sense. However, the Mandal Com-
mission’s claim of 52% of the population
being OBC was based on somewhat flimsy
empirical evidence and if the Census 2011
identifies more than 52% of the popula-
tion as being OBC, this would bolster the
claims for higher representation. Succes-
sive rounds of National Sample Survey
(NSS) have documented the number of in-
dividuals identifying themselves as OBC at
36% in 1999-2000 and at 41% in 2004-05.
Hence it is unlikely that the actual count
will exceed 52%. However there exist
­other, less partisan, arguments for a caste
census. Social background continues to
define privilege and lack thereof in Indian
society resulting in demands for positive
discrimination from marginalised groups.
Unfortunately, these demands are contin-
ually stymied by lack of data. What data
exist from sample surveys continue to
document disparities in education,
­income and standards of living between
different socio-religious communities.
­Table 1 (p 11), based on the India Human
Deve­lopment Survey (IHDS), documents
substantial disparities in different mark-
ers of human development across differ-
ent social groups. This nationally repre-
sentative survey of 41,554 households was
organised by researchers from National
Council of ­Applied Economic Research
(NCAER) and University of M­aryland and
was ­carried out in 2004-05.
The IHDS is not unique in document-
ing these disparities. A vast number of
studies based on NSS, National Family
Health Survey (NFHS) and other sample
surveys have documented inequalities
between scheduled castes (SCs), sched-
uled tribes (STs), OBCs and forward
castes. However, utilisation of survey
statistics is by no means adequate for
e­v idence-based policy design. Survey
data suffer from two shortcomings. First,
even in large surveys such as the NSS,
sample sizes get extremely small once

we begin to compare groups on several
dimensions of disadvantage. For exam-
ple, if we wanted to address the question
of whether Christian dalits are disadvan-
taged like other dalits or whether they
are closer to the general Christian popu-
lation, this question cannot be easily an-
swered due to the small number of
Christian dalits in various surveys. Any
advocacy for affirmative action for Mus-
lim OBCs also suffers from this con-
straint. The only sensible way of collect-
ing data for these smaller communities
is to rely on a nationwide census.
Second, advocacy for refining reserva-
tions is even more affected by lack of recent
data. Castes have been notified as belong-
ing to SC, ST or OBC categories largely (al-
though not solely) based on outdated data
from the 1931 Census. This is particularly
true for castes identified as SC, ST, or OBC in
the early years of reservations. Since a
caste, once notified rarely seems to lose this
status, their subsequent economic prosper-
ity, if any, remains unnoticed. Where disad-
vantage is associated with social distance
and discrimination, such as with dalits or
adivasis, this is not particularly relevant.
Even middle class dalits and adivasis con-
tinue to suffer from disadvantages and so-
cial exclusion (Desai and Kulkarni 2008;
Navsarjan Trust and RFK Center 2010).
However, where disadvantages are associ-
ated with historical exclusion from access
to education or productive resources rather
than active discrimination, as with some of
the OBC communities like the jats, the situ-
ation is more fluid. Some OBC communities
may continue to be economically disadvan-
taged; others may be on par with the
­forward castes. T­able 1 also shows that a
substantial proportion of OBC families have
a college graduate in the household, have
access to salaried jobs and live in homes
with flush toilets. Under these situations,
obtaining accurate data for better targeting
of benefits is crucial.
If inappropriate inclusion of some castes
in reserved category is a problem, exclu-
sion of others from benefits is also prob-
lematic. As Table 1 documents, a substan-
tial proportion of forward caste families
also suffer from socio-economic disadvan-
tages. Whether these disadvantages reflect
social class or are due to some castes being
particularly disadvantaged cannot be
Economic & Political Weekly  EPW   july 17, 2010  vol xlv no 29
COMMENTARY
a­ scertained without better data. Only ac- yielded a list of 7,372 castes. Some of these
curate data reflecting modern Indian reali- are different spellings of the same group,
ty would allow us to ­answer the question Jadav vs Yadav; but others refer to totally
of whether there is greater variation within different but similar sounding groups, e g,
or between castes on various d­imensions Jat vs Jatav.
of material dis­advantage. If there were Given these complexities, it is not sur-
greater variation across castes, this would prising that since independence, the
bolster claims for caste-based affirmative O­ffice of the Registrar General has stead-
action and ­reclassification. fastly avoided collecting detailed caste
data in decennial censuses.
Complexities of Collecting Data
on Caste Collection of Data on SCs/STs
These two imperatives – identifying small- in 2001 Census
er marginalised communities and better While the census does not collect detailed
identification of caste groups for policy data on caste, the 2001 Census did collect
targeting – place tremendous demands on information on SCs and STs. The census
the nature of data collection. First, in enumerators were armed with official
­order to have sufficient information on state-specific lists of SCs/STs and were
small populations groups, we need to col- asked to check whether the caste identifi-
lect the data in census rather than through cation provided by the household matched
household surveys. Second, we need to the list. They were marked as being SC or
enumerate caste details for everyone rath- ST if their caste or tribe was included in
er than just the SC, ST or OBC categories the list. Moreover, only Hindu, Sikh or
and even there we need to get information Buddhists were marked as being SC, re-
at the jati level rather than at the level of gardless of their individual claims to SC
broad population groupings. status. This process follows the official
Collecting data on jatis is hard enough, policy in identifying SC/ST households.
given thousands of jatis inhabiting the However, even here, considerable com-
length and breadth of India; making sense plexities were observed. In some cases,
of these data after collection is even harder. the castes noted by the enumerators on
For example, it is easy for an interviewer to the census forms did not match the caste
ask a respondent about their caste list and the data entry personnel were re-
and write down this response verbatim. quired to check with anthropologists in
Table 1: Social Group Differences in Indicators of Human Development
Mean Males 15-49 Children Males Daliy Income from Homes with
Household with College 8-11 15-59 with Wage and Salaries Flush Toilets
Income Degree or Diploma Who Can Read Salaried Work Males Aged 15-59 (%)
(%) (%) (%) Rural Urban
Forward caste Hindu 72,717 17 71 32 112 243 37
OBC 42,331 8 56 21 77 154 20
Scheduled caste 34,128 5 44 19 69 142 14
Scheduled tribe 32,345 4 46 13 62 180 7
Muslim 44,158 6 45 20 86 114 24
Other religion 1,01,536 16 79 35 147 228 59
All India 47,804 9 54 22 79 173 23
Source: Desai et al (2010) based on India Human Development Survey 2004-05.
However, caste identification can be at a the Social Studies Division of the Office of
very general level or at an extremely fine- the Registrar General of India (RGI) before
grained level. For example, depending deciding on the SC or ST status. The com-
upon her mood, a respondent may say “I plexities of this process and concern that
am Baniya”, “I am Modh” or “I am Dasha mistakes in identification could lead to
Shrimali Modh Banik”. All of these re- under-enumeration of some tribes and
sponses would be correct. If she calls her- lead to premature conclusions about their
self Modh and her cousin calls herself impending extinction undergirds the relu­
Baniya, our computerised analysis would ctance of the RGI for collecting OBC data.
assume they are from different castes. In Moreover, the OBC list is considerably
the IHDS, a sample of 41,554 households longer than the SC and ST lists. Hence the
11
COMMENTARY
burden on the enumerator for flipping
through caste codes could be enormous,
resulting in many enumerators using a
short-cut and simply counting many
households as belonging to forward castes
in order to avoid having to look through
the whole list. Much of the objection to
collection of caste data in census has
c­entred on measurement error in self-­
identification of caste, while measure-
ment error due to deliberate enumerator
misrepresentation has received little at-
tention. However, most experienced sur-
vey designers would tell us that when
work burden can be reduced by misclassi-
fication on the screener question, inter-
viewers are apt to do so. For example, de-
mographic and health surveys find that
when interviewers have to ask a whole se-
ries of questions for children under five,
miraculously the number of six-year olds
increa­ses! In order to avoid this, the infor-
mation must be collected for the whole
population and not just for the target
group. This also serves the broader policy
objectives since the case for refining af-
firmative action targeting requires that
caste information be collected for the en-
tire population and not just for individuals
who fall in the r­eserved category. ed and ease of analysis so that we may ag-
What Would It Take?
If we were to design a full caste census,
what would it look like? Hitherto, the cen-
sus approach has been to take the official
list of castes and tribes falling under the SC
and ST categories and confirm with the re-
spondents whether they belong to one of
these. An alternative to this highly struc-
tured approach is a fully flexible approach
in which the households are asked to pro-
vide their caste identification, enumerators
would write it down fully and then classify
it at the data entry phase. Both of these are
extremely difficult in practice. For a struc-
tured approach we must start with a list of
castes and as of now no such list exists for
the whole population. A flexible approach
has a potential for turning into a classifica-
tion nightmare along the lines discussed
earlier in this article.
Some of these issues were debated at
length in working groups set up in prepa-
ration for the 2001 Census but their delib-
erations were ignored subsequently. One
possible strategy for a caste census may be
12
to adapt the strategy used for the creation surveys sponsored by the government.
of industrial and occupational classifica- This should be the last question in any
tion systems. The occupational classifica- survey so as not to distort other ­responses
tion system provides an interesting illus- and would simply provide one line for
tration. In 2004, a new system of National writing down the full caste name.
Classification of Occupations (NCO-04) ● This list should be augmented using vari-
was developed which contains the follow- ous other sources where caste data are
ing hierarchical structure: Division (10), available such as marriage advertise-
Subdivision (30), Group (116), Family ments, lists of caste associations and
(439), Occupations (2,945). so on. The list can be combined with the
Under this schema, a bidi furnace oper- caste list from the 1931 Census to create
ator would be classified with a code a ­superset.
7,416.45, where the major division is seven ● This broad list can be taken by sociolo-
(craft and related workers), subdivision is gists and anthropologists to come up
74 (craft and related workers excluding with a classification schema following
metal workers, building workers and tex- the occupational classification approach
tile and printing workers), group is 741 outlined above with sub-castes grouped
(food processing and related workers), within broader castes and allowing
family is 7,416 (tobacco preparers), and space for the possibility that the
occupation is 7,416.45 (bidi furnace opera- ­enumerator may encounter a caste
tor). However, if the respondent were to for which no predefined category is
say that he engages in some highly spe- specified.
cialised task within bidi preparation that ● In creation of the caste classification
is not included in the list (for example, system two issues should be kept in
counting and making bidi bundles), he mind. First, caste system should be
could be classified as 7,416.90, tobacco treated independently from religion al-
worker not elsewhere classified. This clas- lowing for the possibility that we will
sification scheme both provides ease of collect sufficient information to identify
classification when data are being collect- Muslim OBCs (or Christian brahmins).
Second, place of residence should be re-
gregate or disaggregate data depending corded to allow for a creation of district
upon the level of detail desired. specific caste list.
What would it take to come up with such ● This caste classification list should be
fine-grained classification system for caste sorted by district so that the enumera-
enumeration? Once again, the processes tors can be given two lists. One list con-
through which occupational and industrial taining about 100 castes most frequent-
systems are developed offer some insights. ly found in each district, and a second
A possible analogous process might be list containing all castes found through-
the following: out India. Given the geographic cluster-
● Obtain a list of jatis residing through ing of castes and tribes in various
the length and breadth of India. This ­regions of India, it seems likely that in
could be done in surveys such as the about 80% of the cases, respondents
NSS, NFHS as well as other unrelated would only need to refer to the district
G.B. Pant Social Science Institute
Jhusi, Allahabad – 211019
ADMISSION IN DOCTORAL PROGRAMME
The Institute invites applications for its doctoral programme from candidates
awarded UGC Junior Research Fellowships (JRF) in Economics, Anthropology,
Sociology and History. Candidates may send their applications by email at
director@gbpssi.org.in or by post by August 14, 2010.
No TA shall be paid but accommodation shall be provided for the day of
the interview.
Director
july 17, 2010  vol xlv no 29  EPW   Economic & Political Weekly

specific list of castes and tribes. For ex-
ample, Garasias will only rarely be
found in West Bengal but frequently in
Gujarat and Rajasthan.
● Once this classification list is developed
it should be tested by incorporation in
large nationwide sample surveys.
● This can then be incorporated in the
2021 Census.
As we look at the complexity of creating
a full list of jatis and subjatis and including
it in an enumeration system, it is clear that
it is unlikely that at this late date any data
collection system can be designed for it
­incor­poration in the 2011 Census that
would do justice to the complexity of I­ndian
social organisation and would meet the
policy demands for data. However, unless
we start thinking about this now, when the
Review of the Rashtriya
Swasthya Bima Yojana
D Narayana
The enrolment of the poor in
the flagship health insurance
scheme, the Rashtriya Swasthya
Bima Yojana in its third year of
operation does not show any
sign of it covering all the poor by
2012. This article estimates the
proportion of the eligible below
the poverty line families
enrolled for the scheme and the
fraction of those hospitalised
who are covered.
D Narayana (narayana@cds.ac.in) is with the
Centre for Development Studies,
Thiruvananthapuram.
Economic & Political Weekly  EPW   july 17, 2010  vol xlv no 29
national attention is centred on the impor-
tance of caste data, it is unlikely that we will
be ready even by 2021. If we start now, it is
not unrealistic that we may be able to obtain
caste information in 2021, 90 years after the
1931 Census when caste data were last col-
lected. Fortunately, by 2021 technological
advances will allow each enumerator to
carry a handheld device or Personal Digital
Assistant with her to avoid her having to
flip through pages and pages of caste list to
find a classification code for the Dasha
Shrimali Modh Banik respondent she is likely
to encounter in Surat but not in Shillong.
References
Anderson, Benedict (1983): Imagined Communities:
Reflections on the Origin and Spread of Nationa­
lism (New York: Verso).
I
ndia is one of the countries with the
lowest government spending on
health; over 80% of the total spending
is by individuals privately, most of it being
out of pocket. For the poor, such large
spending could become a direct entry to
debt trap, or could lead to lack of access.
Introduction of the Universal Health In-
surance Scheme in 2003 was one of the
early steps to alleviate such a burden,
wherein the central government was pay-
ing Rs 100 per household as subsidy to of-
fer a benefit package, with a sum a­ssured
of Rs 30,000. Despite various modifica-
tions, the scheme hardly evinced any re-
sponse, and in 2008, the central govern-
ment launched the Rashtriya Swasthya
Bima Yojana (RSBY).
The RSBY has been designed to provide
health insurance coverage up to Rs 30,000
for the below poverty line (BPL) families
for hospitalisation. Pre-existing condi-
tions are covered from day one and there
is no age limit. Coverage extends to five
members of the family, and they need to
pay only Rs 30 as registration fee with the
premium being paid by the central (75%)
COMMENTARY
Das, Veena, ed. (2003): Handbook of Indian Sociology
(New Delhi: Oxford University Press).
Desai, Sonalde, Amaresh Dubey, B L Joshi, Mitali Sen,
Abusaleh Shariff and Reeve Vanneman (2010): Hu­
man Development in India: Challenges for a Society
in Transition (New Delhi: Oxford University Press).
Desai, Sonalde and Veena Kulkarni (2008): “Changing
Educational Inequalities in India: In the C­ontext of
Affirmative Action”, Demography, 45  (2): 245-70.
Deshpande, Ashwini (2000): “Does Caste Still Define
Disparity? A Look at Inequality in Kerala, India”,
The American Economic Review, 90 (2):322-25.
Dirks, Nicholas (2001): Castes of Mind: Colonialism
and the Making of Modern India (Princeton, NJ:
Princeton University Press).
Govinda, R (2002): “India Education Report” (New
Delhi: Oxford University Press).
Navsarjan Trust and RFK Center (2010): Understand­
ing Untouchability: A Comprehensive Study of
Practices and Conditions in 1589 Villages,
20102010, available from http://www.rfkcenter.
org/untouchability.
Thorat, Sukhdeo and Katherine Newman, ed. (2009):
Blocked by Caste: Economic Discrimination in Mod­
ern India (New Delhi: Oxford University Press).
Zuberi, Tokufu (2001): Thicker Than Blood: How
­Racial Statistics Lie (Minneapolis: University of
Minnesota Press).
and state governments (25%). Every bene-
ficiary family is issued a smart card and all
the empanelled hospitals are IT-­enabled to
ensure smooth flow of data regarding
­service utilisation. The beneficiary has the
choice of public and private hospitals for
cashless treatment and the public hospitals
have the incentive to treat bene­ficiaries, as
under RSBY the money will flow d­irectly to
the concerned hospital.
It is two years since the RSBY has been
launched and a limited evaluation is in
o­rder, which this paper seeks to attempt.
The questions taken up for analysis are the
following: What proportion of the eligible
BPL families has been enrolled for the
scheme? What proportion of the hospitali-
sations is getting covered? What explains
the low coverage of hospitalisations in some
states/districts? Are the data made availa-
ble at the RSBY web site adequate to carry
out a systematic analysis of the scheme?
The article provides a background of
the cost of hospitalisation, proportion of
households getting reimbursements and
the proportion being reimbursed by insur-
ance companies, using information con-
tained in the National Sample Survey
O­rganisation (NSSO) Report No 507
(60/25.0/1) titled, “Morbidity, Healthcare
and Condition of the Aged”. It also takes
up the coverage of the poor under RSBY by
analysing the number of districts covered,
the proportion of poor covered in the dis­
tricts where enrolment has been completed
13